What's new in this version
Broader, more reliable continuity across your AI chats.
- Now works across Claude, ChatGPT, Gemini, Grok, DeepSeek, and Perplexity
- Cleaner objective tracking — your goal updates as the conversation evolves
- More accurate sorting of what's settled, what's new, and what's still open
- Better at reading the full conversation on each platform
- Runs entirely in your browser — nothing leaves your device

v2.4.4 — Continuity that keeps up with your conversation
Prompt Accelerator now tracks your goals, constraints, and decisions across an entire chat — and updates them as the conversation evolves. Change your mind mid-thread ("actually, make it Japan and Seoul") and the running objective updates with you, instead of getting stuck on your first message.
- Captures the full conversation, not just your latest message
- Recognizes when you change the goal and follows the change
- Cleaner separation of what's settled, what's new, and what's still open
- More reliable review across Claude, ChatGPT, Gemini, Grok, DeepSeek, and Perplexity
- Runs entirely in your browser — nothing leaves your device

Version 2.3.2 is a stabilization release focused on improving Advanced panel opening reliability across supported AI providers.

What changed:
- Improved first-click opening behavior for the Advanced review panel.
- Added build provenance diagnostics to help distinguish store builds, local builds, and session state differences.
- Preserved the current continuity review baseline for broader testing.

Known issues:
- Capsule truthfulness is still under active investigation.
- Objective normalization can still fail in some continuity review/export cases.
- Prompt-shell or stale session fragments may still affect exported capsules.
- Rejected directions and unresolved issues may not always be preserved cleanly.
- Perplexity extraction can remain weaker when only retrieved context is available.

This release is intended as a stable open-path baseline while the remaining continuity export issues are addressed separately.

Version 2.3.1 improves Prompt Review reliability and handoff safety.

What changed:
- Cleaner prompt capture on supported AI chat surfaces.
- Better filtering of provider UI text, review debris, and assistant-generated content.
- Stronger preservation of rejected directions, unresolved tensions, governance notes, and invariants.
- Stricter SAFE/UNSAFE handoff checks with clearer reasons when export is blocked.
- Improved review persistence and diagnostics for safer inspection before copying or exporting.

LuxCrypta remains local-first with no backend, telemetry, analytics, or hidden network path.

Version 2.3.0 expands LuxCrypta Prompt Accelerator provider support and continuity diagnostics.

What’s new:
- Added DeepSeek support for structured reasoning continuity workflows.
- Added Perplexity support for retrieval-governed continuity workflows.
- Added provider-specific continuity profiles for DeepSeek and Perplexity.
- Added safeguards so Perplexity retrieved/cited material is treated as Provisional or Quarantine unless explicitly admitted.
- Added provider health diagnostics for surface detection, input detection, toolbar mounting, draft read/write, duplicate guards, and runtime errors.

Changed:
- Updated supported surface permissions for DeepSeek and Perplexity.
- Updated provider handoff routing for ChatGPT, Claude, Gemini, Grok, DeepSeek, and Perplexity.
- Improved export diagnostics to preserve provider profile and retrieval context separately.
- Updated supported-surface documentation and release materials.

Validation:
- Passed typecheck, lint, full test suite, Chromium package build, Firefox package build, source package build, and Firefox web-ext lint.

Version 2.2.2 improves saved continuity artifact quality.

Release notes:

• Improved Workflow, Capsule, and Diagnostic exports with stricter semantic admission filtering.
• Reduced contamination from UI labels, upload chatter, screenshot instructions, copied directive blocks, and handoff scaffolding.
• Portable Workflow exports now rebuild key continuity fields from cleaned admitted state instead of reusing potentially noisy saved fields.
• Recommended next actions now pass through the same admission filtering layer.
• Capsule reconstruction instructions now preserve rejected directions as part of continuity identity.
• Added regression coverage to protect export cleanliness and saved-state hygiene.
• No changes to permissions or supported AI surfaces.

Version 2.2.1 refines the Continuity Review experience for clarity, trust, and submission readiness.

What’s new:
- Improved Review toolbar feedback for Apply, Copy, Save Workflow, and Save Capsule.
- Copy now exports a cleaner, human-readable continuity review for reuse in chats, notes, or documents.
- Save Workflow and Save Capsule now create clearer reusable local artifacts.
- Review columns now use a balanced side-by-side layout with internal scrolling.
- Added stronger draft writeback checks to reduce silent Apply failures.
- Added cross-surface reliability checks for ChatGPT, Claude, Gemini, and Grok adapters.

Also included:
- Stability fixes
- Submission package updates
- Updated Chromium and Firefox builds for version 2.2.1

Version 2.2.0 introduces LuxCrypta’s always-on continuity runtime.

What’s new:
- The default chat-page experience is now quieter: Powered by LuxCrypta + Advanced.
- Compress and Focus are no longer separate user-facing controls.
- Redundancy reduction, objective prioritization, and continuity shaping now run automatically during review flows.
- Advanced now opens a human-readable continuity review first, including Clean Summary, Active Objective, Stable Core, New / Provisional, Open / Unresolved, What Changed, and Recommended Next Actions.
- Raw/internal diagnostics are now secondary and collapsed by default.
- Carry-forward capsules are parsed into readable continuity state instead of being surfaced as raw structure.
- Repeated Advanced/Apply passes are more stable and avoid reintroducing duplicate runtime instructions.
- ChatGPT, Claude, Gemini, and Grok support has been validated for draft read, Advanced review, and Apply writeback.

This release makes Prompt Accelerator feel more like a quiet continuity layer and less like a manual prompt tool, while preserving local-first operation and user-controlled review/apply behavior.

Added first-tier Grok support for the browser extension on grok.com.

What’s new:
- Added Grok as a supported chat surface.
- Added Grok to provider/model targeting alongside ChatGPT, Claude, and Gemini.
- Added Grok-aware prompt adaptation and continuity handoff support.
- Added support for reading, reviewing, and applying drafts on grok.com.
- Added narrow grok.com content-script matching without adding X/Twitter permissions.
- Added regression coverage for Grok surface detection, draft handling, manifest scope, provider routing, and review/apply flow.

Also improved:
- Updated supported-platform documentation and store submission notes.
- Tightened source-package exclusions for cleaner review uploads.
- Rebuilt Chrome, Firefox, and source packages for version 2.1.2.

Version 2.1.1 fixes a duplicate-output issue on ChatGPT pages by removing a legacy response overlay script that could append a flattened second copy of assistant responses.

Also included:
- Keeps ChatGPT response streaming untouched
- Preserves the normal Prompt Accelerator toolbar and review workflow
- No new permissions
- Packaging and Firefox AMO validation cleanup